2 * Local APIC handling, local APIC timers
4 * (c) 1999, 2000 Ingo Molnar <mingo@redhat.com>
7 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
8 * thanks to Eric Gilmore
10 * for testing these extensively.
11 * Maciej W. Rozycki : Various updates and fixes.
12 * Mikael Pettersson : Power Management for UP-APIC.
14 * Mikael Pettersson : PM converted to driver model.
17 #include <linux/init.h>
20 #include <linux/delay.h>
21 #include <linux/bootmem.h>
22 #include <linux/interrupt.h>
23 #include <linux/mc146818rtc.h>
24 #include <linux/kernel_stat.h>
25 #include <linux/sysdev.h>
26 #include <linux/ioport.h>
27 #include <linux/clockchips.h>
28 #include <linux/acpi_pmtmr.h>
29 #include <linux/module.h>
31 #include <asm/atomic.h>
34 #include <asm/mpspec.h>
36 #include <asm/pgalloc.h>
39 #include <asm/proto.h>
40 #include <asm/timex.h>
44 #include <mach_apic.h>
46 static int disable_apic_timer __cpuinitdata;
47 static int apic_calibrate_pmtmr __initdata;
50 /* Local APIC timer works in C2 */
51 int local_apic_timer_c2_ok;
52 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
55 * Debug level, exported for io_apic.c
59 /* Have we found an MP table */
62 static struct resource lapic_resource = {
64 .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
67 static unsigned int calibration_result;
69 static int lapic_next_event(unsigned long delta,
70 struct clock_event_device *evt);
71 static void lapic_timer_setup(enum clock_event_mode mode,
72 struct clock_event_device *evt);
73 static void lapic_timer_broadcast(cpumask_t mask);
74 static void apic_pm_activate(void);
76 static struct clock_event_device lapic_clockevent = {
78 .features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT
79 | CLOCK_EVT_FEAT_C3STOP | CLOCK_EVT_FEAT_DUMMY,
81 .set_mode = lapic_timer_setup,
82 .set_next_event = lapic_next_event,
83 .broadcast = lapic_timer_broadcast,
87 static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
89 static unsigned long apic_phys;
91 unsigned long mp_lapic_addr;
93 unsigned int __cpuinitdata maxcpus = NR_CPUS;
95 * Get the LAPIC version
97 static inline int lapic_get_version(void)
99 return GET_APIC_VERSION(apic_read(APIC_LVR));
103 * Check, if the APIC is integrated or a seperate chip
105 static inline int lapic_is_integrated(void)
111 * Check, whether this is a modern or a first generation APIC
113 static int modern_apic(void)
115 /* AMD systems use old APIC versions, so check the CPU */
116 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
117 boot_cpu_data.x86 >= 0xf)
119 return lapic_get_version() >= 0x14;
122 void xapic_wait_icr_idle(void)
124 while (apic_read(APIC_ICR) & APIC_ICR_BUSY)
128 u32 safe_xapic_wait_icr_idle(void)
135 send_status = apic_read(APIC_ICR) & APIC_ICR_BUSY;
139 } while (timeout++ < 1000);
144 void xapic_icr_write(u32 low, u32 id)
146 apic_write(APIC_ICR2, id << 24);
147 apic_write(APIC_ICR, low);
150 u64 xapic_icr_read(void)
154 icr2 = apic_read(APIC_ICR2);
155 icr1 = apic_read(APIC_ICR);
157 return (icr1 | ((u64)icr2 << 32));
160 static struct apic_ops xapic_ops = {
161 .read = native_apic_mem_read,
162 .write = native_apic_mem_write,
163 .write_atomic = native_apic_mem_write_atomic,
164 .icr_read = xapic_icr_read,
165 .icr_write = xapic_icr_write,
166 .wait_icr_idle = xapic_wait_icr_idle,
167 .safe_wait_icr_idle = safe_xapic_wait_icr_idle,
170 struct apic_ops __read_mostly *apic_ops = &xapic_ops;
172 EXPORT_SYMBOL_GPL(apic_ops);
175 * enable_NMI_through_LVT0 - enable NMI through local vector table 0
177 void __cpuinit enable_NMI_through_LVT0(void)
181 /* unmask and set to NMI */
183 apic_write(APIC_LVT0, v);
187 * lapic_get_maxlvt - get the maximum number of local vector table entries
189 int lapic_get_maxlvt(void)
191 unsigned int v, maxlvt;
193 v = apic_read(APIC_LVR);
194 maxlvt = GET_APIC_MAXLVT(v);
199 * This function sets up the local APIC timer, with a timeout of
200 * 'clocks' APIC bus clock. During calibration we actually call
201 * this function twice on the boot CPU, once with a bogus timeout
202 * value, second time for real. The other (noncalibrating) CPUs
203 * call this function only once, with the real, calibrated value.
205 * We do reads before writes even if unnecessary, to get around the
206 * P5 APIC double write bug.
209 static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
211 unsigned int lvtt_value, tmp_value;
213 lvtt_value = LOCAL_TIMER_VECTOR;
215 lvtt_value |= APIC_LVT_TIMER_PERIODIC;
217 lvtt_value |= APIC_LVT_MASKED;
219 apic_write(APIC_LVTT, lvtt_value);
224 tmp_value = apic_read(APIC_TDCR);
225 apic_write(APIC_TDCR, (tmp_value
226 & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE))
230 apic_write(APIC_TMICT, clocks);
234 * Setup extended LVT, AMD specific (K8, family 10h)
236 * Vector mappings are hard coded. On K8 only offset 0 (APIC500) and
237 * MCE interrupts are supported. Thus MCE offset must be set to 0.
240 #define APIC_EILVT_LVTOFF_MCE 0
241 #define APIC_EILVT_LVTOFF_IBS 1
243 static void setup_APIC_eilvt(u8 lvt_off, u8 vector, u8 msg_type, u8 mask)
245 unsigned long reg = (lvt_off << 4) + APIC_EILVT0;
246 unsigned int v = (mask << 16) | (msg_type << 8) | vector;
251 u8 setup_APIC_eilvt_mce(u8 vector, u8 msg_type, u8 mask)
253 setup_APIC_eilvt(APIC_EILVT_LVTOFF_MCE, vector, msg_type, mask);
254 return APIC_EILVT_LVTOFF_MCE;
257 u8 setup_APIC_eilvt_ibs(u8 vector, u8 msg_type, u8 mask)
259 setup_APIC_eilvt(APIC_EILVT_LVTOFF_IBS, vector, msg_type, mask);
260 return APIC_EILVT_LVTOFF_IBS;
264 * Program the next event, relative to now
266 static int lapic_next_event(unsigned long delta,
267 struct clock_event_device *evt)
269 apic_write(APIC_TMICT, delta);
274 * Setup the lapic timer in periodic or oneshot mode
276 static void lapic_timer_setup(enum clock_event_mode mode,
277 struct clock_event_device *evt)
282 /* Lapic used as dummy for broadcast ? */
283 if (evt->features & CLOCK_EVT_FEAT_DUMMY)
286 local_irq_save(flags);
289 case CLOCK_EVT_MODE_PERIODIC:
290 case CLOCK_EVT_MODE_ONESHOT:
291 __setup_APIC_LVTT(calibration_result,
292 mode != CLOCK_EVT_MODE_PERIODIC, 1);
294 case CLOCK_EVT_MODE_UNUSED:
295 case CLOCK_EVT_MODE_SHUTDOWN:
296 v = apic_read(APIC_LVTT);
297 v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
298 apic_write(APIC_LVTT, v);
300 case CLOCK_EVT_MODE_RESUME:
301 /* Nothing to do here */
305 local_irq_restore(flags);
309 * Local APIC timer broadcast function
311 static void lapic_timer_broadcast(cpumask_t mask)
314 send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
319 * Setup the local APIC timer for this CPU. Copy the initilized values
320 * of the boot CPU and register the clock event in the framework.
322 static void setup_APIC_timer(void)
324 struct clock_event_device *levt = &__get_cpu_var(lapic_events);
326 memcpy(levt, &lapic_clockevent, sizeof(*levt));
327 levt->cpumask = cpumask_of_cpu(smp_processor_id());
329 clockevents_register_device(levt);
333 * In this function we calibrate APIC bus clocks to the external
334 * timer. Unfortunately we cannot use jiffies and the timer irq
335 * to calibrate, since some later bootup code depends on getting
336 * the first irq? Ugh.
338 * We want to do the calibration only once since we
339 * want to have local timer irqs syncron. CPUs connected
340 * by the same APIC bus have the very same bus frequency.
341 * And we want to have irqs off anyways, no accidental
345 #define TICK_COUNT 100000000
347 static void __init calibrate_APIC_clock(void)
349 unsigned apic, apic_start;
350 unsigned long tsc, tsc_start;
356 * Put whatever arbitrary (but long enough) timeout
357 * value into the APIC clock, we just want to get the
358 * counter running for calibration.
360 * No interrupt enable !
362 __setup_APIC_LVTT(250000000, 0, 0);
364 apic_start = apic_read(APIC_TMCCT);
365 #ifdef CONFIG_X86_PM_TIMER
366 if (apic_calibrate_pmtmr && pmtmr_ioport) {
367 pmtimer_wait(5000); /* 5ms wait */
368 apic = apic_read(APIC_TMCCT);
369 result = (apic_start - apic) * 1000L / 5;
376 apic = apic_read(APIC_TMCCT);
378 } while ((tsc - tsc_start) < TICK_COUNT &&
379 (apic_start - apic) < TICK_COUNT);
381 result = (apic_start - apic) * 1000L * tsc_khz /
387 printk(KERN_DEBUG "APIC timer calibration result %d\n", result);
389 printk(KERN_INFO "Detected %d.%03d MHz APIC timer.\n",
390 result / 1000 / 1000, result / 1000 % 1000);
392 /* Calculate the scaled math multiplication factor */
393 lapic_clockevent.mult = div_sc(result, NSEC_PER_SEC,
394 lapic_clockevent.shift);
395 lapic_clockevent.max_delta_ns =
396 clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
397 lapic_clockevent.min_delta_ns =
398 clockevent_delta2ns(0xF, &lapic_clockevent);
400 calibration_result = result / HZ;
404 * Setup the boot APIC
406 * Calibrate and verify the result.
408 void __init setup_boot_APIC_clock(void)
411 * The local apic timer can be disabled via the kernel commandline.
412 * Register the lapic timer as a dummy clock event source on SMP
413 * systems, so the broadcast mechanism is used. On UP systems simply
416 if (disable_apic_timer) {
417 printk(KERN_INFO "Disabling APIC timer\n");
418 /* No broadcast on UP ! */
419 if (num_possible_cpus() > 1) {
420 lapic_clockevent.mult = 1;
426 printk(KERN_INFO "Using local APIC timer interrupts.\n");
427 calibrate_APIC_clock();
430 * Do a sanity check on the APIC calibration result
432 if (calibration_result < (1000000 / HZ)) {
434 "APIC frequency too slow, disabling apic timer\n");
435 /* No broadcast on UP ! */
436 if (num_possible_cpus() > 1)
442 * If nmi_watchdog is set to IO_APIC, we need the
443 * PIT/HPET going. Otherwise register lapic as a dummy
446 if (nmi_watchdog != NMI_IO_APIC)
447 lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
449 printk(KERN_WARNING "APIC timer registered as dummy,"
450 " due to nmi_watchdog=%d!\n", nmi_watchdog);
455 void __cpuinit setup_secondary_APIC_clock(void)
461 * The guts of the apic timer interrupt
463 static void local_apic_timer_interrupt(void)
465 int cpu = smp_processor_id();
466 struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
469 * Normally we should not be here till LAPIC has been initialized but
470 * in some cases like kdump, its possible that there is a pending LAPIC
471 * timer interrupt from previous kernel's context and is delivered in
472 * new kernel the moment interrupts are enabled.
474 * Interrupts are enabled early and LAPIC is setup much later, hence
475 * its possible that when we get here evt->event_handler is NULL.
476 * Check for event_handler being NULL and discard the interrupt as
479 if (!evt->event_handler) {
481 "Spurious LAPIC timer interrupt on cpu %d\n", cpu);
483 lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt);
488 * the NMI deadlock-detector uses this.
490 add_pda(apic_timer_irqs, 1);
492 evt->event_handler(evt);
496 * Local APIC timer interrupt. This is the most natural way for doing
497 * local interrupts, but local timer interrupts can be emulated by
498 * broadcast interrupts too. [in case the hw doesn't support APIC timers]
500 * [ if a single-CPU system runs an SMP kernel then we call the local
501 * interrupt as well. Thus we cannot inline the local irq ... ]
503 void smp_apic_timer_interrupt(struct pt_regs *regs)
505 struct pt_regs *old_regs = set_irq_regs(regs);
508 * NOTE! We'd better ACK the irq immediately,
509 * because timer handling can be slow.
513 * update_process_times() expects us to have done irq_enter().
514 * Besides, if we don't timer interrupts ignore the global
515 * interrupt lock, which is the WrongThing (tm) to do.
519 local_apic_timer_interrupt();
521 set_irq_regs(old_regs);
524 int setup_profiling_timer(unsigned int multiplier)
531 * Local APIC start and shutdown
535 * clear_local_APIC - shutdown the local APIC
537 * This is called, when a CPU is disabled and before rebooting, so the state of
538 * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
539 * leftovers during boot.
541 void clear_local_APIC(void)
546 /* APIC hasn't been mapped yet */
550 maxlvt = lapic_get_maxlvt();
552 * Masking an LVT entry can trigger a local APIC error
553 * if the vector is zero. Mask LVTERR first to prevent this.
556 v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
557 apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
560 * Careful: we have to set masks only first to deassert
561 * any level-triggered sources.
563 v = apic_read(APIC_LVTT);
564 apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
565 v = apic_read(APIC_LVT0);
566 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
567 v = apic_read(APIC_LVT1);
568 apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
570 v = apic_read(APIC_LVTPC);
571 apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
575 * Clean APIC state for other OSs:
577 apic_write(APIC_LVTT, APIC_LVT_MASKED);
578 apic_write(APIC_LVT0, APIC_LVT_MASKED);
579 apic_write(APIC_LVT1, APIC_LVT_MASKED);
581 apic_write(APIC_LVTERR, APIC_LVT_MASKED);
583 apic_write(APIC_LVTPC, APIC_LVT_MASKED);
584 apic_write(APIC_ESR, 0);
589 * disable_local_APIC - clear and disable the local APIC
591 void disable_local_APIC(void)
598 * Disable APIC (implies clearing of registers
601 value = apic_read(APIC_SPIV);
602 value &= ~APIC_SPIV_APIC_ENABLED;
603 apic_write(APIC_SPIV, value);
606 void lapic_shutdown(void)
613 local_irq_save(flags);
615 disable_local_APIC();
617 local_irq_restore(flags);
621 * This is to verify that we're looking at a real local APIC.
622 * Check these against your board if the CPUs aren't getting
623 * started for no apparent reason.
625 int __init verify_local_APIC(void)
627 unsigned int reg0, reg1;
630 * The version register is read-only in a real APIC.
632 reg0 = apic_read(APIC_LVR);
633 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg0);
634 apic_write(APIC_LVR, reg0 ^ APIC_LVR_MASK);
635 reg1 = apic_read(APIC_LVR);
636 apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg1);
639 * The two version reads above should print the same
640 * numbers. If the second one is different, then we
641 * poke at a non-APIC.
647 * Check if the version looks reasonably.
649 reg1 = GET_APIC_VERSION(reg0);
650 if (reg1 == 0x00 || reg1 == 0xff)
652 reg1 = lapic_get_maxlvt();
653 if (reg1 < 0x02 || reg1 == 0xff)
657 * The ID register is read/write in a real APIC.
659 reg0 = apic_read(APIC_ID);
660 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
661 apic_write(APIC_ID, reg0 ^ APIC_ID_MASK);
662 reg1 = apic_read(APIC_ID);
663 apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
664 apic_write(APIC_ID, reg0);
665 if (reg1 != (reg0 ^ APIC_ID_MASK))
669 * The next two are just to see if we have sane values.
670 * They're only really relevant if we're in Virtual Wire
671 * compatibility mode, but most boxes are anymore.
673 reg0 = apic_read(APIC_LVT0);
674 apic_printk(APIC_DEBUG, "Getting LVT0: %x\n", reg0);
675 reg1 = apic_read(APIC_LVT1);
676 apic_printk(APIC_DEBUG, "Getting LVT1: %x\n", reg1);
682 * sync_Arb_IDs - synchronize APIC bus arbitration IDs
684 void __init sync_Arb_IDs(void)
686 /* Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 */
693 apic_wait_icr_idle();
695 apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
696 apic_write(APIC_ICR, APIC_DEST_ALLINC | APIC_INT_LEVELTRIG
701 * An initial setup of the virtual wire mode.
703 void __init init_bsp_APIC(void)
708 * Don't do the setup now if we have a SMP BIOS as the
709 * through-I/O-APIC virtual wire mode might be active.
711 if (smp_found_config || !cpu_has_apic)
714 value = apic_read(APIC_LVR);
717 * Do not trust the local APIC being empty at bootup.
724 value = apic_read(APIC_SPIV);
725 value &= ~APIC_VECTOR_MASK;
726 value |= APIC_SPIV_APIC_ENABLED;
727 value |= APIC_SPIV_FOCUS_DISABLED;
728 value |= SPURIOUS_APIC_VECTOR;
729 apic_write(APIC_SPIV, value);
732 * Set up the virtual wire mode.
734 apic_write(APIC_LVT0, APIC_DM_EXTINT);
736 apic_write(APIC_LVT1, value);
740 * setup_local_APIC - setup the local APIC
742 void __cpuinit setup_local_APIC(void)
748 value = apic_read(APIC_LVR);
750 BUILD_BUG_ON((SPURIOUS_APIC_VECTOR & 0x0f) != 0x0f);
753 * Double-check whether this APIC is really registered.
754 * This is meaningless in clustered apic mode, so we skip it.
756 if (!apic_id_registered())
760 * Intel recommends to set DFR, LDR and TPR before enabling
761 * an APIC. See e.g. "AP-388 82489DX User's Manual" (Intel
762 * document number 292116). So here it goes...
767 * Set Task Priority to 'accept all'. We never change this
770 value = apic_read(APIC_TASKPRI);
771 value &= ~APIC_TPRI_MASK;
772 apic_write(APIC_TASKPRI, value);
775 * After a crash, we no longer service the interrupts and a pending
776 * interrupt from previous kernel might still have ISR bit set.
778 * Most probably by now CPU has serviced that pending interrupt and
779 * it might not have done the ack_APIC_irq() because it thought,
780 * interrupt came from i8259 as ExtInt. LAPIC did not get EOI so it
781 * does not clear the ISR bit and cpu thinks it has already serivced
782 * the interrupt. Hence a vector might get locked. It was noticed
783 * for timer irq (vector 0x31). Issue an extra EOI to clear ISR.
785 for (i = APIC_ISR_NR - 1; i >= 0; i--) {
786 value = apic_read(APIC_ISR + i*0x10);
787 for (j = 31; j >= 0; j--) {
794 * Now that we are all set up, enable the APIC
796 value = apic_read(APIC_SPIV);
797 value &= ~APIC_VECTOR_MASK;
801 value |= APIC_SPIV_APIC_ENABLED;
803 /* We always use processor focus */
806 * Set spurious IRQ vector
808 value |= SPURIOUS_APIC_VECTOR;
809 apic_write(APIC_SPIV, value);
814 * set up through-local-APIC on the BP's LINT0. This is not
815 * strictly necessary in pure symmetric-IO mode, but sometimes
816 * we delegate interrupts to the 8259A.
819 * TODO: set up through-local-APIC from through-I/O-APIC? --macro
821 value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
822 if (!smp_processor_id() && !value) {
823 value = APIC_DM_EXTINT;
824 apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n",
827 value = APIC_DM_EXTINT | APIC_LVT_MASKED;
828 apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n",
831 apic_write(APIC_LVT0, value);
834 * only the BP should see the LINT1 NMI signal, obviously.
836 if (!smp_processor_id())
839 value = APIC_DM_NMI | APIC_LVT_MASKED;
840 apic_write(APIC_LVT1, value);
844 static void __cpuinit lapic_setup_esr(void)
846 unsigned maxlvt = lapic_get_maxlvt();
848 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR);
850 * spec says clear errors after enabling vector.
853 apic_write(APIC_ESR, 0);
856 void __cpuinit end_local_APIC_setup(void)
859 setup_apic_nmi_watchdog(NULL);
864 * Detect and enable local APICs on non-SMP boards.
865 * Original code written by Keir Fraser.
866 * On AMD64 we trust the BIOS - if it says no APIC it is likely
867 * not correctly set up (usually the APIC timer won't work etc.)
869 static int __init detect_init_APIC(void)
872 printk(KERN_INFO "No local APIC present\n");
876 mp_lapic_addr = APIC_DEFAULT_PHYS_BASE;
877 boot_cpu_physical_apicid = 0;
881 void __init early_init_lapic_mapping(void)
883 unsigned long phys_addr;
886 * If no local APIC can be found then go out
887 * : it means there is no mpatable and MADT
889 if (!smp_found_config)
892 phys_addr = mp_lapic_addr;
894 set_fixmap_nocache(FIX_APIC_BASE, phys_addr);
895 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
896 APIC_BASE, phys_addr);
899 * Fetch the APIC ID of the BSP in case we have a
900 * default configuration (or the MP table is broken).
902 boot_cpu_physical_apicid = GET_APIC_ID(read_apic_id());
906 * init_apic_mappings - initialize APIC mappings
908 void __init init_apic_mappings(void)
911 * If no local APIC can be found then set up a fake all
912 * zeroes page to simulate the local APIC and another
913 * one for the IO-APIC.
915 if (!smp_found_config && detect_init_APIC()) {
916 apic_phys = (unsigned long) alloc_bootmem_pages(PAGE_SIZE);
917 apic_phys = __pa(apic_phys);
919 apic_phys = mp_lapic_addr;
921 set_fixmap_nocache(FIX_APIC_BASE, apic_phys);
922 apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
923 APIC_BASE, apic_phys);
926 * Fetch the APIC ID of the BSP in case we have a
927 * default configuration (or the MP table is broken).
929 boot_cpu_physical_apicid = GET_APIC_ID(read_apic_id());
933 * This initializes the IO-APIC and APIC hardware if this is
936 int __init APIC_init_uniprocessor(void)
939 printk(KERN_INFO "Apic disabled\n");
944 printk(KERN_INFO "Apic disabled by BIOS\n");
952 physid_set_mask_of_physid(boot_cpu_physical_apicid, &phys_cpu_present_map);
953 apic_write(APIC_ID, SET_APIC_ID(boot_cpu_physical_apicid));
958 * Now enable IO-APICs, actually call clear_IO_APIC
959 * We need clear_IO_APIC before enabling vector on BP
961 if (!skip_ioapic_setup && nr_ioapics)
964 if (!smp_found_config || skip_ioapic_setup || !nr_ioapics)
965 localise_nmi_watchdog();
966 end_local_APIC_setup();
968 if (smp_found_config && !skip_ioapic_setup && nr_ioapics)
972 setup_boot_APIC_clock();
973 check_nmi_watchdog();
978 * Local APIC interrupts
982 * This interrupt should _never_ happen with our APIC/SMP architecture
984 asmlinkage void smp_spurious_interrupt(void)
990 * Check if this really is a spurious interrupt and ACK it
991 * if it is a vectored one. Just in case...
992 * Spurious interrupts should not be ACKed.
994 v = apic_read(APIC_ISR + ((SPURIOUS_APIC_VECTOR & ~0x1f) >> 1));
995 if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
998 add_pda(irq_spurious_count, 1);
1003 * This interrupt should never happen with our APIC/SMP architecture
1005 asmlinkage void smp_error_interrupt(void)
1011 /* First tickle the hardware, only then report what went on. -- REW */
1012 v = apic_read(APIC_ESR);
1013 apic_write(APIC_ESR, 0);
1014 v1 = apic_read(APIC_ESR);
1016 atomic_inc(&irq_err_count);
1018 /* Here is what the APIC error bits mean:
1021 2: Send accept error
1022 3: Receive accept error
1024 5: Send illegal vector
1025 6: Received illegal vector
1026 7: Illegal register address
1028 printk(KERN_DEBUG "APIC error on CPU%d: %02x(%02x)\n",
1029 smp_processor_id(), v , v1);
1034 * * connect_bsp_APIC - attach the APIC to the interrupt system
1036 void __init connect_bsp_APIC(void)
1041 void disconnect_bsp_APIC(int virt_wire_setup)
1043 /* Go back to Virtual Wire compatibility mode */
1044 unsigned long value;
1046 /* For the spurious interrupt use vector F, and enable it */
1047 value = apic_read(APIC_SPIV);
1048 value &= ~APIC_VECTOR_MASK;
1049 value |= APIC_SPIV_APIC_ENABLED;
1051 apic_write(APIC_SPIV, value);
1053 if (!virt_wire_setup) {
1055 * For LVT0 make it edge triggered, active high,
1056 * external and enabled
1058 value = apic_read(APIC_LVT0);
1059 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1060 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1061 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1062 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1063 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
1064 apic_write(APIC_LVT0, value);
1067 apic_write(APIC_LVT0, APIC_LVT_MASKED);
1070 /* For LVT1 make it edge triggered, active high, nmi and enabled */
1071 value = apic_read(APIC_LVT1);
1072 value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
1073 APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
1074 APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
1075 value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
1076 value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
1077 apic_write(APIC_LVT1, value);
1080 void __cpuinit generic_processor_info(int apicid, int version)
1085 if (num_processors >= NR_CPUS) {
1086 printk(KERN_WARNING "WARNING: NR_CPUS limit of %i reached."
1087 " Processor ignored.\n", NR_CPUS);
1091 if (num_processors >= maxcpus) {
1092 printk(KERN_WARNING "WARNING: maxcpus limit of %i reached."
1093 " Processor ignored.\n", maxcpus);
1098 cpus_complement(tmp_map, cpu_present_map);
1099 cpu = first_cpu(tmp_map);
1101 physid_set(apicid, phys_cpu_present_map);
1102 if (apicid == boot_cpu_physical_apicid) {
1104 * x86_bios_cpu_apicid is required to have processors listed
1105 * in same order as logical cpu numbers. Hence the first
1106 * entry is BSP, and so on.
1110 if (apicid > max_physical_apicid)
1111 max_physical_apicid = apicid;
1113 /* are we being called early in kernel startup? */
1114 if (early_per_cpu_ptr(x86_cpu_to_apicid)) {
1115 u16 *cpu_to_apicid = early_per_cpu_ptr(x86_cpu_to_apicid);
1116 u16 *bios_cpu_apicid = early_per_cpu_ptr(x86_bios_cpu_apicid);
1118 cpu_to_apicid[cpu] = apicid;
1119 bios_cpu_apicid[cpu] = apicid;
1121 per_cpu(x86_cpu_to_apicid, cpu) = apicid;
1122 per_cpu(x86_bios_cpu_apicid, cpu) = apicid;
1125 cpu_set(cpu, cpu_possible_map);
1126 cpu_set(cpu, cpu_present_map);
1129 int hard_smp_processor_id(void)
1131 return read_apic_id();
1140 /* 'active' is true if the local APIC was enabled by us and
1141 not the BIOS; this signifies that we are also responsible
1142 for disabling it before entering apm/acpi suspend */
1144 /* r/w apic fields */
1145 unsigned int apic_id;
1146 unsigned int apic_taskpri;
1147 unsigned int apic_ldr;
1148 unsigned int apic_dfr;
1149 unsigned int apic_spiv;
1150 unsigned int apic_lvtt;
1151 unsigned int apic_lvtpc;
1152 unsigned int apic_lvt0;
1153 unsigned int apic_lvt1;
1154 unsigned int apic_lvterr;
1155 unsigned int apic_tmict;
1156 unsigned int apic_tdcr;
1157 unsigned int apic_thmr;
1160 static int lapic_suspend(struct sys_device *dev, pm_message_t state)
1162 unsigned long flags;
1165 if (!apic_pm_state.active)
1168 maxlvt = lapic_get_maxlvt();
1170 apic_pm_state.apic_id = apic_read(APIC_ID);
1171 apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
1172 apic_pm_state.apic_ldr = apic_read(APIC_LDR);
1173 apic_pm_state.apic_dfr = apic_read(APIC_DFR);
1174 apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
1175 apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
1177 apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
1178 apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
1179 apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
1180 apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
1181 apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
1182 apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
1183 #ifdef CONFIG_X86_MCE_INTEL
1185 apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
1187 local_irq_save(flags);
1188 disable_local_APIC();
1189 local_irq_restore(flags);
1193 static int lapic_resume(struct sys_device *dev)
1196 unsigned long flags;
1199 if (!apic_pm_state.active)
1202 maxlvt = lapic_get_maxlvt();
1204 local_irq_save(flags);
1205 rdmsr(MSR_IA32_APICBASE, l, h);
1206 l &= ~MSR_IA32_APICBASE_BASE;
1207 l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
1208 wrmsr(MSR_IA32_APICBASE, l, h);
1209 apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
1210 apic_write(APIC_ID, apic_pm_state.apic_id);
1211 apic_write(APIC_DFR, apic_pm_state.apic_dfr);
1212 apic_write(APIC_LDR, apic_pm_state.apic_ldr);
1213 apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
1214 apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
1215 apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
1216 apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
1217 #ifdef CONFIG_X86_MCE_INTEL
1219 apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
1222 apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
1223 apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
1224 apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
1225 apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
1226 apic_write(APIC_ESR, 0);
1227 apic_read(APIC_ESR);
1228 apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
1229 apic_write(APIC_ESR, 0);
1230 apic_read(APIC_ESR);
1231 local_irq_restore(flags);
1235 static struct sysdev_class lapic_sysclass = {
1237 .resume = lapic_resume,
1238 .suspend = lapic_suspend,
1241 static struct sys_device device_lapic = {
1243 .cls = &lapic_sysclass,
1246 static void __cpuinit apic_pm_activate(void)
1248 apic_pm_state.active = 1;
1251 static int __init init_lapic_sysfs(void)
1257 /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
1259 error = sysdev_class_register(&lapic_sysclass);
1261 error = sysdev_register(&device_lapic);
1264 device_initcall(init_lapic_sysfs);
1266 #else /* CONFIG_PM */
1268 static void apic_pm_activate(void) { }
1270 #endif /* CONFIG_PM */
1273 * apic_is_clustered_box() -- Check if we can expect good TSC
1275 * Thus far, the major user of this is IBM's Summit2 series:
1277 * Clustered boxes may have unsynced TSC problems if they are
1278 * multi-chassis. Use available data to take a good guess.
1279 * If in doubt, go HPET.
1281 __cpuinit int apic_is_clustered_box(void)
1283 int i, clusters, zeros;
1285 u16 *bios_cpu_apicid;
1286 DECLARE_BITMAP(clustermap, NUM_APIC_CLUSTERS);
1289 * there is not this kind of box with AMD CPU yet.
1290 * Some AMD box with quadcore cpu and 8 sockets apicid
1291 * will be [4, 0x23] or [8, 0x27] could be thought to
1292 * vsmp box still need checking...
1294 if ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && !is_vsmp_box())
1297 bios_cpu_apicid = early_per_cpu_ptr(x86_bios_cpu_apicid);
1298 bitmap_zero(clustermap, NUM_APIC_CLUSTERS);
1300 for (i = 0; i < NR_CPUS; i++) {
1301 /* are we being called early in kernel startup? */
1302 if (bios_cpu_apicid) {
1303 id = bios_cpu_apicid[i];
1305 else if (i < nr_cpu_ids) {
1307 id = per_cpu(x86_bios_cpu_apicid, i);
1314 if (id != BAD_APICID)
1315 __set_bit(APIC_CLUSTERID(id), clustermap);
1318 /* Problem: Partially populated chassis may not have CPUs in some of
1319 * the APIC clusters they have been allocated. Only present CPUs have
1320 * x86_bios_cpu_apicid entries, thus causing zeroes in the bitmap.
1321 * Since clusters are allocated sequentially, count zeros only if
1322 * they are bounded by ones.
1326 for (i = 0; i < NUM_APIC_CLUSTERS; i++) {
1327 if (test_bit(i, clustermap)) {
1328 clusters += 1 + zeros;
1334 /* ScaleMP vSMPowered boxes have one cluster per board and TSCs are
1335 * not guaranteed to be synced between boards
1337 if (is_vsmp_box() && clusters > 1)
1341 * If clusters > 2, then should be multi-chassis.
1342 * May have to revisit this when multi-core + hyperthreaded CPUs come
1343 * out, but AFAIK this will work even for them.
1345 return (clusters > 2);
1349 * APIC command line parameters
1351 static int __init apic_set_verbosity(char *str)
1354 skip_ioapic_setup = 0;
1358 if (strcmp("debug", str) == 0)
1359 apic_verbosity = APIC_DEBUG;
1360 else if (strcmp("verbose", str) == 0)
1361 apic_verbosity = APIC_VERBOSE;
1363 printk(KERN_WARNING "APIC Verbosity level %s not recognised"
1364 " use apic=verbose or apic=debug\n", str);
1370 early_param("apic", apic_set_verbosity);
1372 static __init int setup_disableapic(char *str)
1375 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
1378 early_param("disableapic", setup_disableapic);
1380 /* same as disableapic, for compatibility */
1381 static __init int setup_nolapic(char *str)
1383 return setup_disableapic(str);
1385 early_param("nolapic", setup_nolapic);
1387 static int __init parse_lapic_timer_c2_ok(char *arg)
1389 local_apic_timer_c2_ok = 1;
1392 early_param("lapic_timer_c2_ok", parse_lapic_timer_c2_ok);
1394 static __init int setup_noapictimer(char *str)
1396 if (str[0] != ' ' && str[0] != 0)
1398 disable_apic_timer = 1;
1401 __setup("noapictimer", setup_noapictimer);
1403 static __init int setup_apicpmtimer(char *s)
1405 apic_calibrate_pmtmr = 1;
1409 __setup("apicpmtimer", setup_apicpmtimer);
1411 static int __init lapic_insert_resource(void)
1416 /* Put local APIC into the resource map. */
1417 lapic_resource.start = apic_phys;
1418 lapic_resource.end = lapic_resource.start + PAGE_SIZE - 1;
1419 insert_resource(&iomem_resource, &lapic_resource);
1425 * need call insert after e820_reserve_resources()
1426 * that is using request_resource
1428 late_initcall(lapic_insert_resource);